Content uploaded by Florence Rosey
Author content
All content in this area was uploaded by Florence Rosey on Mar 20, 2018
Content may be subject to copyright.
Author's personal copy
Accident
Analysis
and
Prevention
45 (2012) 91–
98
Contents
lists
available
at
SciVerse
ScienceDirect
Accident
Analysis
and
Prevention
j
ourna
l
h
o
mepage:
www.elsevier.com/locate/aap
The
impact
of
perceptual
treatments
on
driver’s
behavior:
From
driving
simulator
studies
to
field
tests—First
results
Jean-Michel
Auberleta,∗,
Florence
Roseyb,
Franc¸
oise
Anceauxc,
Sébastien
Aubind,
Patrice
Briande,
Marie-Pierre
Pacauxc,
Patrick
Plainchaultd
aUniversité
Paris
Est,
IFSTTAR-LEPSIS,
F-75732
Paris,
France
bCETE
Normandie
Centre
–
DESGI
&
ERA
34,
10
chemin
de
la
poudrière,
F-76121
Le
Grand
Quevilly
Cedex,
France
cLAMIH
–
UMR
CNRS
8530,
Université
de
Valenciennes
et
du
Hainaut-Cambrésis
–
Le
Mont
Houy,
F-59313
Valenciennes
Cedex
9,
France
dCER-ESEO,
4
rue
Merlet
de
la
Boulaye
–
BP30926,
F-49009
Angers
Cedex
01,
France
eCETE
de
l’Ouest,
LRPC
Angers,
23
Avenue
de
l’Amiral
Chauvin
–
BP
69,
F-49136
Les
ponts-de-Cé
Cedex,
France
a
r
t
i
c
l
e
i
n
f
o
Article
history:
Received
4
June
2010
Received
in
revised
form
12
November
2011
Accepted
26
November
2011
Keywords:
Driving
simulator
Validation
Trajectory
monitoring
station
Perceptual
treatments
Lateral
position
a
b
s
t
r
a
c
t
Our
study
focused
on
the
lateral
position
of
drivers
in
relation
to
risk
on
rural
crest
vertical
curves,
using
a
field
site
proposed
by
a
local
operator
of
the
French
road
network
(Conseil
Général
de
Maine-et-Loire,
49).
The
final
goal
was
to
test
one
road
treatment
on
this
field
site.
The
study
consisted
of
three
stages.
The
first,
using
driving
simulators,
selected
two
perceptual
treatments
(i.e.,
rumble
strips
on
both
sides
of
the
centerline
and
sealed
shoulders)
from
five
that
were
tested
in
order
to
help
drivers
maintain
lateral
control
when
driving
on
crest
vertical
curves.
The
rumble
strips
were
installed
first
on
the
field
site.
The
second
stage
was
to
develop
a
diagnostic
device
specifically
in
order
to
evaluate,
on
the
field
site,
the
impact
of
a
perceptual
treatment
on
the
driver’s
performance
(i.e.,
lateral
position).
This
diagnostic
device
was
installed
in
the
field
upstream
and
downstream
of
the
target
crest
vertical
curve.
The
third
stage
was
to
collect
the
data
during
two
periods,
before
and
after
the
centerline
rumble
strips
were
installed.
We
then
compared
the
results
obtained
in
the
field
study
with
those
from
the
driving
simulator
studies.
The
comparison
showed
that,
as
in
the
simulator
studies,
the
centerline
rumble
strips
on
the
crest
vertical
curve
affected
lateral
positions,
causing
the
participants
to
drive
closer
to
the
center
of
the
lane.
Finally,
the
results
showed
the
usefulness
of
driving
simulators
in
the
road
design
process.
© 2011 Elsevier Ltd. All rights reserved.
1.
Introduction
In
Europe,
80%
of
all
accidents
on
rural
roads
can
be
divided
into
four
categories:
single-vehicle
accidents
(e.g.,
running
off
the
road,
head-on
collisions
with
an
obstacle),
head-on
collisions
between
two
vehicles,
collisions
at
intersections
and
accidents
involving
vul-
nerable
road
users.
Single-vehicle
and
head-on
collisions
(which
relate
to
trajectory
control)
account
for
48%
of
all
crashes
(OECD,
1999).
In
France,
40%
of
road
fatalities
occur
in
single-vehicle
acci-
dents
without
a
third
party.
In
90%
of
cases,
they
involve
a
loss
of
control
generally
leading
to
a
head-on
collision
with
a
rigid
obstacle
at
the
side
of
the
road
(ONISR,
2010).
The
proportion
of
deaths
due
to
this
kind
of
collisions
(i.e.,
12%)
has
hardly
changed
in
the
last
10
∗Corresponding
author.
Tel.:
+33
(0)1
4043
6569.
E-mail
addresses:
jean-michel.auberlet@ifsttar.fr
(J.-M.
Auberlet),
florence.rosey@wanadoo.fr
(F.
Rosey),
Francoise.Anceaux@univ-valenciennes.fr
(F.
Anceaux),
sebastien.aubin@eseo.fr
(S.
Aubin),
Patrice.Briand@developpement-
durable.gouv.fr
(P.
Briand),
marie-pierre.lemoine@univ-valenciennes.fr
(M.-P.
Pacaux),
patrick.plainchault@eseo.fr
(P.
Plainchault).
years
(ONISR,
2010).
Head-on
collisions
are
responsible
for
half
the
deaths
in
collisions
between
two
vehicles
and
mostly
occur
dur-
ing
overtaking
maneuvers
(ONISR,
2010).
Head-on
collisions
also
account
for
10%
of
all
injury
accidents
and
20%
of
all
deaths
(ONISR,
2010).
Single-vehicle
accidents
account
for
21%
of
injury
accidents
and
40%
of
deaths
(ONISR,
2010).
Head-on
collisions
with
a
fixed
obstacle
account
for
67%
of
injury
accidents
and
86%
of
deaths
(from
ONISR,
2010).
In
addition,
poor
lateral
positioning
is
one
of
the
primary
factors
leading
to
crashes
(RISER,
2006)
and
human
error
is
estimated
to
play
a
part
in
around
90%
of
crashes
(Dewar
and
Olson,
2002;
Wegman,
2007).
These
failures
could
result
from
a
false
perception
of
the
road
layout
and/or
environment,
which
has
been
identified
as
a
contributing
factor
to
about
30%
of
crashes
(O’Cinneide,
1998;
Rumar,
1985).
Indeed,
the
information
provided
by
the
road
and
road
environment
is
essential
for
the
driver
to
be
able
to
modulate
driving
control
parameters
and
avoid
risky
behav-
ior
(Saad,
2002;
Theeuwes
and
Godthelp,
1995).
In
this
context,
we
have
determined
which
perceptual
treatment
could
most
effec-
tively
improve
trajectory
control,
more
specifically
lateral
position
control
(our
subject)
on
crest
vertical
curves
(CVCs).
The
final
goal
was
to
test
one
road
treatment
on
a
field
site
with
a
crest
vertical
0001-4575/$
–
see
front
matter ©
2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.aap.2011.11.020
Author's personal copy
92 J.-M.
Auberlet
et
al.
/
Accident
Analysis
and
Prevention
45 (2012) 91–
98
curve.
Nevertheless,
CVCs
interfere
with
trajectory
control
because
they
conceal
the
long-range
visual
information
needed
to
predict
the
path
of
the
road
ahead
and
to
anticipate
future
events
(e.g.,
Rumar
and
Marsh,
1998),
which
allow
better
lane
keeping
(e.g.,
Summala,
1998).
Furthermore,
on
a
CVC
on
a
narrow
road
(i.e.,
3
m)
the
temporary
loss
of
visibility
prevents
the
driver
from
seeing
an
oncoming
vehicle.
In
this
situation,
if
the
driver
is
influenced
by
the
roadway
environment
and
drives
close
to
the
centerline
(in
refer-
ence
to
Blana
and
Golias,
2002;
Harms,
1993;
van
Driel
et
al.,
2004),
there
is
a
danger
he/she
will
collide
with
the
oncoming
vehicle.
Fur-
thermore,
the
sudden
appearance
of
an
on-coming
vehicle
can
lead
to
over-correction
due
to
surprise
and
a
run-off-road
crash.
Run-
off-road
crashes
can
also
result
from
driving
close
to
shoulders.
At
excessive
speeds,
these
types
of
over-correction
can
lead
to
a
fatal
crash
(e.g.,
a
collision
or
run-off-road).
In
this
context,
we
have
conducted
a
three-stage
study
with
two
main
goals.
The
first
was
to
verify
that
the
use
of
percep-
tual
treatment
on
CVCs
improves
lateral
position
control
and
more
precisely
that
the
perceptual
treatment
(selected
after
driving
sim-
ulator
studies)
made
drivers
move
toward
the
center
of
the
lane
when
driving
on
the
field
site
with
a
CVC
as
it
had
in
the
driving
simulator
study.
The
second
purpose
was
to
estimate
the
useful-
ness
and
the
validity
of
driving
simulators
for
evaluating
traffic
engineering
issues
such
as
roadway
design
and
traffic
management.
The
value
of
driving
simulators
in
the
highway
design
process
has
already
been
confirmed
by
Keith
et
al.
(2005).
Their
benefits
are
of
several
types,
facilitating
experimental
control
and
data
collec-
tion,
increasing
efficiency
and
safety,
and
reducing
cost
(Bella,
2005,
2008).
They
are
useful
for
testing
roadway
delineations,
which
can
be
expensive
in
the
field
(Molino
et
al.,
2005).
Furthermore,
driving
simulators
provide
a
safe,
inexpensive
and
ethical
facility,
which
appears
to
be
an
ideal
environment
for
testing
perceptual
treat-
ments
(e.g.,
Godley
et
al.,
1999;
Kaptein
et
al.,
1995,
1996).
The
first
stage
of
our
study
consisted
of
choosing
the
percep-
tual
treatment
which
was
to
be
used
on
the
real
site.
Thus,
using
a
fixed-base
driving
simulator,
Rosey
et
al.
(2008)
examined
how
four
perceptual
road
treatments
affected
lateral
position
because
of
their
ability
to
provide
additional
visual
cues
about
the
road
alignment
which
may
allow
the
driver
to
plan
and
execute
steer-
ing
and
speed
control.
The
four
experimental
treatments
were
selected
from
the
literature
on
perceptual
countermeasures
and
in
collaboration
with
road
safety
experts
from
a
French
Engineering
Centre
(Centre
d’Étude
Techniques
de
l’Équipement
“Normandie-
Centre”,
Cété-NC).
These
experts
were
involved
in
the
study
from
an
early
stage.
The
four
perceptual
treatments
tested
were
a
painted
centerline,
post
delineators,
rumble
strips
on
both
sides
of
the
cen-
terline
and
sealed
shoulders.
The
rumble
strips
on
both
sides
of
the
centerline
(CRS)
and
sealed
shoulders
(SS)
were
determined
to
be
the
most
effective.
A
second
study
using
a
motion-base
driv-
ing
simulator
(in
addition
to
the
fixed-base
facility)
has
confirmed
that
these
two
perceptual
treatments
encourage
drivers
to
move
toward
the
center
of
their
lane
(Auberlet
et
al.,
2009).
In
order
to
verify
that
the
CRS
and
the
SS
had
the
same
impact
on
lateral
position
control
on
the
real
CVC
as
on
the
simulated
CVC:
1.
We
developed
a
specific
diagnostic
device
to
track
the
trajectory
of
individual
vehicles
on
the
section
of
road
with
the
CVC
at
the
real
site
(Aubin
et
al.,
2008).
2.
We
then
installed
the
device
and
the
rumble
strips
at
the
real
site,
and
collected
the
trajectories
of
the
individual
vehicles.
Due
to
multiple
factors
(e.g.,
the
cost
of
the
installing
the
treatment),
we
decided,
in
collaboration
with
the
local
authorities
to
test
the
rumble
strips
first.
Thus
the
field
study
described
in
this
paper
focused
on
the
impact
of
the
rumble
strips
on
trajectories.
This
paper
begins
with
a
concise
description
of
the
driving
simulator
studies
that
were
conducted
to
choose
the
perceptual
treatment
to
be
evaluated
on
the
real
road.
The
specifically
devel-
oped
diagnostic
device
is
then
described.
The
results
of
the
data
analysis
are
presented
next.
The
paper
ends
with
a
discussion
about
the
positive
impacts
of
the
evaluated
perceptual
treatment
(i.e.
the
rumble
strips),
and
more
generally
about
the
use
of
the
driving
simulators
in
the
road
design
process.
2.
Materials
and
methods
2.1.
Driving
simulator
study:
choice
of
perceptual
treatment(s)
The
goal
of
the
driving
simulator
study
was
to
assess
the
impact
of
four
pre-selected
perceptual
treatments
(i.e.
painted
centerlines,
post
delineators,
rumble
strips
on
both
sides
of
the
centerline
and
sealed
shoulders)
on
lateral
control
when
driving
on
a
CVC
in
order
to
test
them
on
a
real
CVC.
Full
details
are
given
in
Rosey
et
al.
(2008).
The
pre-selection
process
involved
a
number
of
stages:
(1)
a
review
of
the
existing
perceptual
countermeasures
(PCMs)
used
on
roads
and
in
driving
simulators
(e.g.,
Anund
et
al.,
2008;
Horberry
et
al.,
2006),
(2)
early
collaboration
with
road
safety
experts
from
a
French
Engineering
Centre
(Centre
d’Étude
Techniques
de
l’Équipement
“Normandie-Centre”,
Cété-NC).
This
study
showed
that
rumble
strips
on
both
sides
of
the
centerline
(Centerline
Rum-
ble
Strips
– CRS)
and
Sealed
Shoulders
(SS)
are
effective
means
of
encouraging
drivers
to
keep
closer
to
the
center
of
their
lane
(Rosey
et
al.,
2008).
More
specially,
the
results
showed
that
CRS
and
SS
focus
the
driver’s
attention
on
the
lane
both
immediately
after
entering
the
treated
CVC
zone
(i.e.,
pre-test
hill
section)
and
throughout
it
(i.e.,
test
hill
section).
Furthermore,
the
results
show
that
the
control
treatment
(i.e.,
actual
delineation
marking),
painted
centerline
and
post
delineators
had
little
effect
on
lateral
position
compared
to
the
CRS
and
SS
treatments
(Rosey
et
al.,
2008).
We
shall
now
recall
and
summarize
the
study
and
the
results.
2.1.1.
Participants
Participants
with
full
French
driving
licenses
(i.e.,
neither
learn-
ers
or
persons
with
restricted
licenses)
were
recruited.
They
were
required
to
have
normal
or
corrected
vision.
The
size
of
the
sam-
ples
was
42
participants
(16
women
and
26
men),
between
22
and
58
years
old
(average
age
38.6;
SD
=
10.83).
Their
average
driving
experience
was
19
years
and
they
drove
on
average
12,373
km
per
year.
2.1.2.
Apparatus
The
study
was
conducted
using
the
INRETS-MSIS
SIM2driving
simulator
(Fig.
1),
which
is
an
interactive
fixed-base
driving
sim-
ulator
with
a
complete
PSA-Citroen
Xantia
car
(height:
1380
mm,
width:
1755
mm).
The
driving
simulator
was
positioned
in
front
of
three
angled
projection
surfaces.
The
entire
projection
image
produced
a
150◦
(horizontal)
×
45◦(vertical)
forward
view
of
the
simulated
road-
way
from
the
driver’s
position,
at
resolutions
of
1024
×
768
pixels
(SIM2).
The
control
devices
were
the
steering
wheel,
manual
gear-
box
and
pedals
(brake,
accelerator
and
clutch)
of
the
complete
car.
The
dynamic
vehicle
model
used
in
the
driving
simulator
is
based
on
the
ARHMM
(Advanced
Road
Handling
Multi-body
Model),
which
was
developed
jointly
in
the
1990s
by
INRETS,
PSA
and
Renault
in
the
framework
of
the
SARA
consortium.
The
driving
simulator
provided
haptic
feedback
from
the
steer-
ing
wheel.
Loudspeakers
inside
the
cars
and
a
sub-woofer
in
front
of
the
car
provided
realistic
engine
and
road
noises.
Speakers
around
the
car
created
Doppler
effects
to
simulate
the
noise
of
on-coming
Author's personal copy
J.-M.
Auberlet
et
al.
/
Accident
Analysis
and
Prevention
45 (2012) 91–
98 93
Fig.
1.
The
INRETS-MSIS
SIM2fixed-base
driving
simulator.
traffic
on
both
the
driving
simulators.
Auditory
feedback
was
pro-
vided
when
the
simulated
vehicle’s
wheels
crossed
the
rumble
strips
on
both
sides
of
the
centerline
(CRS)
or
the
sealed
shoul-
der
(SS).
In
the
case
of
these
two
treatments
the
noise
informed
the
participants
when
they
drove
over
them.
2.1.3.
Simulator
scenario
The
roadway
geometry
depicted
in
the
simulations
was
a
virtual
reconstruction
of
a
rural
two-lane
road
(D961)
in
the
Maine-et-
Loire
(Department
49,
France).
It
was
6
m
wide
with
a
speed
limit
of
90
km/h
and
no
road
signs
around
the
target
CVC.
Based
on
the
real
topographic
layout,
this
3D
road
scene
included
a
3
km
straight
section
with
two
CVCs,
the
first
of
which
concealed
the
second
(the
target
CVC).
The
lane
widths,
road
mark-
ings,
sight
distances,
and
other
geometrical
characteristics
were
incorporated
into
the
simulation.
The
driving
conditions
were
day-
light,
clear
vision
and
full
friction.
There
was
light
oncoming
traffic
to
show
the
participants
that
they
could
encounter
other
cars.
As
suggested
by
Lewis-Evans
and
Charlton
(2006),
we
chose
not
to
simulate
heavy
traffic
or
lead
vehicles
to
avoid
influencing
the
par-
ticipants’
speed
and
lateral
positioning.
Fig.
2
shows
a
photograph
of
the
actual
CVC
(left)
and
an
image
of
the
same
CVC
as
it
appeared
in
the
simulation
(right).
The
experiment
dealt
with
the
perceptual
treatments
around
the
target
CVC,
where
visibility
would
be
reduced.
In
order
to
ana-
lyze
the
impact
of
the
perceptual
treatments,
we
defined
a
“test
hill”
section,
which
was
the
zone
in
which
the
treatments
were
used.
This
consisted
of
the
150
m
immediately
before
and
immedi-
ately
after
the
apex
of
the
CVC.
The
four
evaluated
treatments
were:
painted
centerline
(PC),
post
delineators
(PoD),
rumble
strips
on
both
sides
of
the
centerline
(CRS)
and
sealed
shoulders
(SS)
(Fig.
3).
We
also
set
up
a
control,
called
the
control
treatment
(CT).
One
3D
database
was
built
for
each
treatment
(i.e.,
CT,
PC,
PoD,
CRS
and
SS).
For
the
“painted
centerline”
condition,
the
delineation
mark-
ings
were
deleted
before
or
after
the
test
hill
section
(i.e.,
the
crest
vertical
curve).
2.1.4.
Data
collection
Although
the
simulator
collected
a
number
of
parameters,
we
only
processed
lateral
position.
The
lateral
position
measurements
were
continuously
recorded
with
a
sampling
frequency
of
60
Hz.
Since
our
study
was
designed
to
analyze
vehicle
trajectory,
we
used
lane
position
as
a
performance
indicator.
There
were
two
reasons
for
this:
lane
position
provides
a
measure
of
driving
per-
formance
that
describes
the
safety-relevance
of
changes
in
driving
behavior
(McGehee
et
al.,
2004)
and
this
indicator
enables
us
to
determine
the
alerting
effect
of
road
treatments
(Anund
et
al.,
2008;
Horberry
et
al.,
2006).
The
lateral
position
is
defined
as
the
location
of
the
vehicle’s
longitudinal
axis
relative
to
a
longitudi-
nal
road
reference
system
(Porter
et
al.,
2004).
In
our
study,
the
lateral
position
corresponded
to
the
distance
(in
mm)
from
the
vehicle
centroid
to
the
roadway
centerline.
A
lane
position
of
0
mm
would
mean
the
vehicle
was
exactly
straddling
the
roadway
cen-
terline.
In
order
to
analyze
lateral
position
changes,
road
sections
were
defined
both
in
the
vicinity
of
the
CVC
and
distant
from
it
to
pro-
vide
test
sections
for
the
treatments
and
neutral
sections.
We
chose
150-m-long
test
sections,
which
corresponded
to
the
distance
trav-
eled
in
6
s
(i.e.,
5
s
+
one
extra
second)
at
a
theoretical
speed
of
90
km/h
(i.e.,
the
speed
limit
on
this
type
of
rural
road).
In
the
study,
four
road
sections
were
analyzed:
(1)
the
reference
section
which
was
used
to
identify
the
vehicle’s
lateral
position
without
any
treatment;
(2)
the
pre-test
hill
section
which
began
150
m
before
the
beginning
of
the
treatment;
(3)
the
test
hill
section
which
consisted
of
the
300
m
where
the
target
CVC
was
treated
(i.e.,
150
m
on
each
side
of
the
apex
of
the
target
CVC);
and
(4)
the
post-test
section
which
was
the
150
m
after
the
treatment
ended.
During
the
trials
“subjective
data”
concerning
the
participants
(e.g.,
driving
habits,
age,
gender),
their
perception
of
the
situation
and
the
treatments,
as
well
as
their
representation
of
the
driving
activity
were
collected
by
means
of
a
number
of
questionnaires.
In
particular,
immediately
after
each
driving
run
(there
were
five
in
all),
the
participants
completed
a
short
questionnaire
to
gauge
the
acceptability
of
the
treatment
with
regard
to
comfort
and
safety.
These
subjective
data
were
used
to
estimate
any
potential
discrep-
ancies
between
the
participants’
feelings
and
the
objective
data
recorded.
2.1.5.
Statistics
analysis
The
lateral
positions
were
measured
several
times
for
each
par-
ticipant
on
the
four
road
sections
(i.e.,
the
control
section,
the
pre-test
hill,
the
test
hill,
and
the
post-test
hill)
and
under
the
five
conditions
(control
treatment
–
CT,
painted
centerline
–
PC,
post
delineators
–
PoD,
rumble
strips
on
both
sides
of
the
cen-
terline
–
CRS
and
sealed
shoulders
–
SS).
The
lateral
position
data
were
then
analyzed
using
repeated
measures
ANOVA
with
the
perceptual
treatment
(control
treatment,
painted
centerline,
post
delineators,
rumble
strips
on
both
sides
of
the
centerline,
and
sealed
shoulders)
and
the
influence
zone
(reference
sec-
tion,
pre-test
hill,
test-hill,
and
post-test
hill)
as
within-subject
factors.
Before
the
repeated
measures
ANOVAs
were
performed
all
the
data
used
in
the
statistical
analysis
were
subjected
to
the
Kolmogorov–Smirnov
test,
which
showed
that
all
the
data
were
normally
distributed.
This
test
was
not
sufficient,
however,
because
the
F
test
is
robust
to
violations
of
the
multivariate
nor-
mal
assumption,
but
not
to
the
sphericity
assumption
(Lewis,
1993).
When
the
sphericity
assumption
is
violated
(i.e.,
in
this
case
when
the
Mauchley
test
was
significant)
adjustments
were
made
to
the
ANOVA
results
using
the
Geisser–Greenhouse
epsilon,
Author's personal copy
94 J.-M.
Auberlet
et
al.
/
Accident
Analysis
and
Prevention
45 (2012) 91–
98
Fig.
2.
Photograph
of
one
section
of
the
real
road
(left)
and
the
virtual
reconstruction
(right).
which
provides
an
F
test
using
a
much
more
stringent
criterion
(Geisser
and
Greenhouse,
1958).
Thus,
the
decision
about
whether
an
F
test
was
significant
was
based
on
the
Geisser–Greenhouse
epsilon.
Where
the
within-subject
variables
violated
the
sphericity
assumption,
we
have
reported
the
Geisser–Greenhouse
probabil-
ities.
Each
repeated
measures
ANOVA
was
followed
by
post
hoc
Newman–Keuls
test.
The
threshold
for
statistical
significance
was
set
at
p
=
0.05.
2.2.
Field
study:
diagnostic
device
development
and
verification
of
the
CRS
treatment
impacts
2.2.1.
Apparatus
2.2.1.1.
The
specifically
developed
diagnostic
device.
A
new
proto-
type
device
was
developed
that
consisted
of
FBG
sensors,
new
resistive
sensors
and
inductive
loop
detectors.
It
was
set
up
on
an
830
m
section
of
the
RD961
rural
road.
This
trajectory
monitoring
Fig.
3.
The
two
experimental
perceptual
treatments
as
depicted
in
the
simulation.
The
width
of
the
sealed
shoulder
was
1
m.
The
ochre
color
was
chosen
to
avoid
confusion
with
the
red
and
green
used
for
cycle
paths
in
France.
(For
interpretation
of
the
references
to
color
in
this
figure
legend,
the
reader
is
referred
to
the
web
version
of
the
article.)
Author's personal copy
J.-M.
Auberlet
et
al.
/
Accident
Analysis
and
Prevention
45 (2012) 91–
98 95
Fig.
4.
Configuration
of
two
inductive
loop
sensors.
station
was
jointly
developed
by
the
French
laboratory
CER-ESEO
and
the
French
Engineering
Centre
Cété-Ouest.
It
aims
to
follow
the
vehicles
individually
by
recording
their
characteristics
(speed,
lat-
eral
position,
type).
All
the
data
were
synchronized
and
recorded
on
a
server
then
transferred
by
GPRS
to
the
laboratories.
The
FBG
sensors
and
the
resistive
sensors
were
prototypes
(Plainchault
and
Aubin,
2009)
and
broke
down
after
a
few
days.
We
were
therefore
only
able
to
use
the
inductive
loop
data.
2.2.1.2.
Inductive
loop
detectors.
This
is
the
most
common
type
of
sensor
in
traffic
management
applications.
The
wire
loop
is
excited
by
a
signal
ranging
in
frequency
from
10
kHz
to
200
kHz.
The
induc-
tance
of
the
loop
decreases
when
a
vehicle
stops
on
it
or
passes
over
it.
This
decrease
in
inductance
increases
the
oscillation
frequency
and
causes
the
electronic
unit
to
send
a
pulse
to
the
controller,
indicating
the
presence
or
passage
of
a
vehicle.
The
single
wire
is
wound
to
give
a
few
turns.
Like
the
two
other
kinds
of
sensors,
it
is
a
subsurface
system,
installed
less
than
100
mm
under
the
sur-
face.
Cété-Ouest
developed
an
innovative
way
of
using
inductive
loop
detectors
to
measure
lateral
position
(Fig.
4)
(Le
Bastard
and
Briand,
2010).
This
configuration
enabled
us
to
log
the
speed
and
lateral
position
of
the
vehicles.
2.2.2.
Perceptual
treatment
and
data
collection
2.2.2.1.
Installing
the
rumble
strips.
The
trajectory
monitoring
sta-
tion
was
set
up
in
February
2007.
In
December
2008,
the
rumble
strips
were
installed
on
the
central
axis
of
the
road
100
m
before
and
50
m
after
the
apex
of
the
CVC.
The
rumble
strips
were
150
mm
long,
45
mm
wide,
11
mm
high
and
longitudinally
40
cm
apart.
Fig.
5
shows
the
installed
rumble
strips
and
the
trajectory
monitoring
station
on
the
apex
of
the
CVC.
2.2.2.2.
Objective
data.
The
objective
data
were
only
recorded
by
the
10
pairs
of
inductive
loop
detectors,
located
at
5
points
and
on
each
lane
on
the
road
section.
Thus,
speed
and
lateral
positions
were
recorded
by
the
five
pairs
of
inductive
loop
detectors
on
one
lane
in
order
to
measure
the
individual
trajectories
(Fig.
6).
Furthermore,
in
this
study,
since
the
sensors
were
embedded
in
the
roadway,
the
drivers
were
unaware
that
they
were
subjects
in
an
experiment.
Fig.
5.
Photograph
showing
the
design
of
the
trajectory
monitoring
stations.
The
photo
shows
the
three
data
collection
systems
(i.e.,
fiber
Bragg
grating
sensors,
resis-
tive
sensors
and
inductive
loop
detectors)
and
the
installation
of
the
rumble
strips.
The
rumble
strips
(150
mm
long,
45
mm
wide,
11
mm
high
and
longitudinally
40
cm
apart)
were
installed
on
the
central
axis
of
the
road
100
m
before
and
50
m
after
the
apex
of
the
CVC.
In
addition,
in
order
to
compare
driver
trajectories
before
and
after
the
rumble
strips
were
installed,
we
had
to
select
two
similar
periods,
one
between
February
2007
and
November
2008,
and
the
other
between
January
2009
and
March
2009.
With
regard
to
the
temperature,
we
selected
2
weeks
in
March
2008,
1
week
in
January
2009
and
1
week
in
February
2009.
During
these
periods,
the
traffic
flow
was
almost
2000
vehicles/lane/day,
of
which
10%
were
lorries.
Almost
25%
of
the
drivers
were
local
regular
drivers.
2.2.2.3.
Subjective
data.
As
in
the
driving
simulator
experiment,
driver
data
were
collected:
before
the
CRS
was
installed
(Before-set
up,
Before-Su),
and
after
(After-set
up,
After-Su).
In
Before-Su,
40
participants
were
asked
to
fill
in
some
questionnaires
after
driving
through
the
CVC
site.
In
After-Su,
20
participants
were
asked
to
fill
in
the
same
questionnaires
as
in
Before-Su.
The
drivers/participants
were
stopped
about
500
m
after
the
apex
of
the
CVC.
The
trials
were
held
over
2
days,
in
September
2008
(Before-Su),
and
in
March
2009
(After-Su).
3.
Results
The
results
of
the
driving
simulator
study
are
reported
first,
and
the
results
of
the
field
study
second.
For
the
driving
simulator
study,
we
have
only
given
a
detailed
account
of
the
results
for
the
percep-
tual
treatments
(the
CRS
and
the
SS)
on
the
CVC
(i.e.,
the
test-hill
section).
The
reason
was
twofold:
(1)
the
aim
of
the
study
was
to
verify
that
the
impact
of
the
CRS
and
SS
perceptual
treatments
on
the
simulated
CVC
was
similar
to
those
on
the
real
CVC
and
(2)
the
field
study
mainly
focused
on
the
section
before
the
apex
of
the
CVC.
Fig.
6.
View
of
the
road
profile
with
the
five
test
sections
(S1–S5)
used
in
the
dynamic
driving
simulator
study
and
the
positions
of
the
five
pairs
of
sensor
loops
(b1–b5)
used
in
field
study.
Author's personal copy
96 J.-M.
Auberlet
et
al.
/
Accident
Analysis
and
Prevention
45 (2012) 91–
98
3.1.
Driving
simulator
study
3.1.1.
Objective
data
The
repeated
measures
ANOVA
computed
for
the
effects
of
the
perceptual
treatments
revealed
a
reliable
perceptual
treatment
effect
for
the
pre-test
hill
(PrTH,
F(3.10,127.25)
=
4.57,
p
<
0.001,
the
test
hill
(TH,
F(4,164)
=
8.76,
p
<
0.0001)
and
the
post-test
hill
(PoTH,
F(4,164)
=
2.99,
p
<
0.02),
but
not
for
the
reference
section
(RS,
F(3.32,136.16)
=
2.10,
p
=
0.09).
For
the
test
hill,
post
hoc
analyses
indicated
that
the
participants
drove
closer
to
the
center
of
the
road
in
the
case
of
the
reference
road
(RR),
the
painted
centerline
(PC)
and
the
post
delineators
(PoD)
than
in
the
case
of
the
rumble
strips
on
both
sides
of
the
center-
line
(CRS*RR,
p
<
0.0001;
CRS*PC,
p
<
0.02;
CRS*PoD,
p
<
0.0001)
and
the
sealed
shoulders
(SS*RR,
p
<
0.0001;
SS*PC,
p
<
0.0001;
SS*PoD,
p
<
0.0003).
Apart
from
this
they
drove
with
similar
trajectories
in
the
case
of
the
RR
and
PC
(p
=
0.12),
the
RR
and
PoD
(p
=
0.75),
the
PC
and
PoD
(p
=
0.13)
and,
the
CRS
and
SS
(p
=
0.74).
To
summarize,
the
participants
followed
two
types
of
trajectory
on
the
CVC.
Some
drove
closer
to
center
of
the
road
in
the
case
of
the
reference
road
(RR),
the
painted
centerline
(PC),
and
the
post
delineators
(PoD);
and
others
drove
closer
to
the
center
of
the
lane
in
the
case
of
the
rumble
strips
on
both
sides
of
the
centerline
(CRS)
and
the
sealed
shoulders
(SS).
More
precisely,
the
CRS
altered
the
lateral
position
by
about
15
cm
and
the
SS
by
about
14
cm.
These
results
were
similar
to
those
obtained
by
Hatfield
et
al.
(2009)
who
emphasized
the
benefits
of
the
CRS
treatment
in
terms
of
lane-keeping
and
visibility.
Furthermore,
our
results
stress
the
effectiveness
of
CRS
with
regard
to
lateral
position,
a
finding
that
is
shared
by
a
recent
driving
simulator
study
on
sleepiness
(Anund
et
al.,
2008).
3.1.2.
Subjective
data
The
participants
had
a
stronger
feeling
of
safety
and
comfort
with
the
CRS
and
SS
perceptual
treatments
than
without.
Further-
more,
these
perceptual
treatments
reduced
the
surprise
effect
of
the
CVC.
The
subjects
reported
a
preference
for
the
SS
treatment,
but
they
did
not
remember
perceiving
the
treatments:
31%
of
the
partici-
pants
did
not
recall
the
CRS
treatments
and
43%
of
the
participants
did
not
recall
the
SS
treatments.
Moreover,
the
trajectory
patterns
of
these
participants
were
similar
to
those
of
participants
who
did
recall
the
treatments.
3.1.3.
Conclusion
Based
on
the
results
from
the
driving
simulator
studies
(Rosey
et
al.,
2008;
Auberlet
et
al.,
2009),
in
collaboration
with
the
local
operator,
we
decided
to
evaluate
the
two
perceptual
treatments
(CRS,
SS)
at
the
field
site.
This
provides
a
way
of
approaching
the
issue
of
the
reliability
of
the
results
from
driving
simulators
and
their
transferability
to
the
real
environment.
We
shall
now
report
our
first
findings
concerning
the
CRS.
One
difficulty
is
that
in
France,
CRS
are
not
used
on
rural
roads,
so
we
had
to
find
a
similar
treatment.
We
chose
VNTP
which
is
an
audio
tactile
profiled
road
marking
enhancing
night
time
vis-
ibility
during
rain
or
wet
conditions.
In
our
case,
the
VNTP
used
was
a
thermoplastic
highway
marking
system
that
gives
drivers
a
sensory
(vibration)
warning.
This
treatment
was
installed
on
the
centerline
road
marking.
Furthermore,
in
the
absence
of
any
rules
about
how
to
install
the
VNTP
in
this
position,
in
agreement
with
the
local
operator,
we
decided
to
shorten
the
length
of
the
treat-
ment
after
the
apex
of
the
CVC
to
avoid
disturbing
riders
and
drivers
who
wanted
to
overtake.
This
decision
still
allows
us
to
evaluate
the
treatment
since
we
mainly
focused
on
the
section
before
the
apex
of
the
CVC.
3.2.
Field
study
3.2.1.
Objective
data
The
analysis
deals
with
the
two
two-week
periods,
one
in
2008
without
the
rumble
strips
(i.e.,
Before-Su),
and
one
in
2009
with
the
rumble
strips
installed
(i.e.,
After-Su)
on
the
field
site.
The
first
stage
of
processing
was
to
identify
the
vehicles
that
did
not
belong
to
pla-
toons
(lorries
are
not
considered
in
this
paper).
Thus,
we
identified
three
driving
modes:
•following
mode:
the
vehicle
followed
a
lead
vehicle
(i.e.,
the
head-
way
was
less
than
5
s),
with
no
oncoming
traffic;
•free
mode:
the
vehicle
had
no
interactions
with
the
vehicle
in
front
of
it
(i.e.,
the
headway
was
greater
than
5
s),
with
no
oncom-
ing
traffic;
•isolated
mode:
the
vehicle
was
alone
with
no
vehicle
in
front
of
it
and
no
oncoming
traffic
on
the
totality
of
the
830
m
of
the
section.
We
also
identified
four
sets
of
weather
conditions:
daytime
and
night-time,
and
dry
or
rainy.
In
the
case
of
dry
daytime
conditions,
the
VNTP
had
a
significant
impact
on
the
drivers’
lateral
positions,
which
moved
around
8.5
cm
to
the
right.
The
results
were
similar
for
three
other
types
of
weather
conditions
–
with
the
VNTP
the
drivers
shifted
between
8.5
cm
and
11
cm
away
from
the
road
axis
to
the
right.
Table
1
reports
the
results
for
the
“dry
day-time”
and
“dry
night-
time”
groups
at
location
b3
(i.e.,
on
the
apex
of
the
CVC).
It
should
be
noted
that
the
difference
in
the
number
of
vehicles
between
the
two
periods
is
because
the
daytime
is
between
3
and
4
h
shorter
in
January
than
in
March.
Two
paired
t-test
were
conducted
on
the
“dry
day-time”
date
for
the
isolated
vehicle
driving
mode,
in
order
to
have
the
same
con-
ditions
as
the
driving
simulator
studies:
one
of
the
tests
compared
mean
lateral
position
according
the
section
(i.e.,
b1,
b2,
b3)
for
each
period
(Before-Su
and
After-Su),
while
the
other
compared
mean
lateral
position
for
the
same
periods.
The
statistical
analyses
of
the
lateral
position
show
that:
-
As
with
the
driving
simulator
for
the
two
periods,
(1)
the
par-
ticipants
drove
closer
to
the
center
of
the
road
when
climbing
the
CVC
(i.e.,
between
b1
and
b2
the
sections
before
b3,
respec-
tively,
Before-Su:
t39
=
2.37,
p
<
.05;
After-Su:
t19
=
9.02,
p
<
.05)
and
(2)
they
shifted
away
from
it
near
the
apex
of
the
CVC
(respec-
tively,
Before-Su:
b2
vs
b3
=
−7.52,
p
<
0.01;
b1
vs
b3:
t39
=
−2.67,
p
<
0.02
and
After-Su:
b2
vs
b3:
t19
=
−31.56,
p
<
0.01;
b1vs
b3:
t19
=
−24.33,
p
<
0.01).
-
The
trajectories
were
different,
the
participants
drove
closer
to
the
center
of
the
lane
After-Su
than
Before-Su
(respectively,
b1:
t60
=
4.34,
p
<
0.05;
b2:
t60
=
10.65,
p
<
0.02
and
b3:
t60
=
14.24,
p
<
0.01).
3.2.2.
Conclusion
The
preliminary
analysis
of
the
participants’
objective
data
revealed
similar
driving
patterns
in
the
trials
conducted
on
the
driving
simulator
and
the
test
site
under
the
control
condition
(Before-Su)
and
with
the
CRS
(After-Su).
Otherwise,
our
analysis
of
the
subjective
data
shows
an
inter-
esting
lack
of
effects.
Indeed,
behaviors
(speed
and
lateral
position)
do
not
appear
to
be
influenced
by
the
driver’s
characteristics
(age,
gender).
Furthermore,
they
are
not
linked
to
the
driver’s
subjec-
tive
evaluation
of
the
situation:
for
example,
all
the
correlations
between
lateral
position
and
the
answers
to
questions
relating
to
the
perception
or
adequacy
of
the
treatments
and
others
relating
to
the
difficulty
or
hazardous
nature
of
the
situation
are
not
sig-
nificant.
The
same
applies
to
correlations
between
lateral
position
Author's personal copy
J.-M.
Auberlet
et
al.
/
Accident
Analysis
and
Prevention
45 (2012) 91–
98 97
Table
1
Summary
of
the
means
and
standard
deviations
of
lateral
position
(cm),
of
“day
without
rain”
and
of
“night
without
rain”
for
the
three
driving
modes
on
the
field
crest
vertical
curve.
The
lateral
position
corresponded
to
the
distance
of
the
vehicle
centroid
from
the
lane
axis
(and
not
the
road
axis
as
in
driving
simulator
studies).
A
negative
distance
mean
represents
a
subject
that
drove
to
the
left
of
the
lane
(i.e.,
closer
to
the
road
axis)
and
a
positive
distance
mean
represents
a
subject
that
drove
to
the
right
of
the
lane
(i.e.,
drove
closer
to
the
shoulder).
Driving
modes Conditions
Day
without
rain
Night
without
rain
Number
of
vehicles Average
(cm)
Standard
deviation
(cm)
Number
of
vehicles
Average
(cm)
Standard
deviation
(cm)
Following
without
crossing
Before
2026
−7.21
20.06
260
−16.2
22.26
After
1303
0.72
19.73
431
−6.29
21.91
Free
without
crossing
Before
4766
−1.93
19.12
1185
−15.78
21.93
After
3527
6.29
19.2
1507
−7.64
19.98
Isolated Before 202 −4.6
20.41
305 −21.26
23.05
After
167
4.95
19.85
279
−11.71
18.27
and
different
attitudes
and
representations,
for
example
auto-
estimated
control
of
the
situation
and
adaptive
and
cooperative
behaviors.
So,
the
effects
of
the
treatments
appear
to
be
indepen-
dent
of
drivers’
characteristics,
habits
and
attitudes.
4.
Conclusion
and
outlook
The
context
of
our
study
is
the
need
to
inform
drivers
more
effec-
tively
about
the
risk
of
losing
control
on
rural
roads.
It
focused
on
the
risk
related
to
lateral
position
on
crest
vertical
curves
(CVCs)
on
rural
roads.
Trials
of
five
techniques
conducted
on
a
fixed-base
driving
simulator
led
us
to
choose
two
perceptual
treatments,
i.e.
Centerline
Rumble
Strips
(CRS)
and
Sealed
Shoulders
(SS)
which
were
then
tested
on
a
real
crest
vertical
curve.
A
motion-base
driving
simulator
study
had
confirmed
the
impact
of
these
two
per-
ceptual
treatments
on
lateral
position
(Auberlet
et
al.,
2009).
Like
the
driving
simulator
studies,
the
field
study
showed
that
the
CRS
had
a
significant
impact
on
the
drivers’
lateral
position
on
a
CVC,
with
participants
driving
closer
to
the
center
of
the
lane
(i.e.,
they
shifted
away
from
the
road
axis
to
the
right).
The
next
stage
of
field
study
will
be
to
evaluate
the
impact
of
the
SS
(sealed
shoulders).
The
field
study
highlighted
the
positive
impacts
of
a
raised
ther-
moplastic
centerline
marking
(rumble
strips)
on
the
driver’s
lateral
position
control.
The
comparison
of
the
results
of
the
driving
simulator
study
with
those
of
the
field
study
showed
their
reliability.
The
relative
behavioral
validity
found
in
our
study
confirms
that
the
driving
simulator
is
a
reliable
tool
for
the
analysis
of
driver
behavior
in
road-
way
design
in
the
sense
that
it
allows
researchers
to
determine
in
advance,
in
a
safe
and
controlled
environment,
how
treatments
or
environmental
modifications
will
affect
driver
behaviors.
The
use
of
a
driving
simulator
improves
our
understanding
of
the
phys-
ical
space
that
is
being
designed
without
endangering
drivers,
and
allows
us
to
make
a
quantitative
evaluation
of
the
“safety”
as
opposed
to
the
hazardousness
of
alternative
designs,
which
makes
it
possible
to
select
the
most
appropriate
type
of
technique
before
undertaking
detailed
design
studies.
Nevertheless,
caution
is
required
when
extrapolating
the
results
to
the
real-life
driv-
ing
environment.
For
example,
while
we
obtained
a
good
level
of
relative
validity
(i.e.,
similarity
between
the
effects
of
different
vari-
ations
in
the
driving
situation)
between
the
driving
simulator
study
and
the
field
study;
the
impact
of
the
CRS
(centerline
rumble
strip)
was
greater
in
virtual
reality
than
in
real-life.
On
the
field
site
the
drivers
moved
between
8.5
cm
and
11
cm
away
from
the
road
axis,
compared
with
15.8
cm
in
virtual
reality.
Other
studies
(e.g.,
Philip
et
al.,
2005;
Rosey
et
al.,
2009)
comparing
results
between
real
driv-
ing
and
simulated
driving,
showed
effects
with
higher
amplitudes
(in
the
context
of
12
h
of
driving
for
Philip
et
al.,
2005
and
nar-
rower
lane
width
for
Rosey
et
al.,
2009)
in
simulated
environments
than
in
real
environments.
Other
studies
have
shown
higher
driving
speeds
on
simulated
roads
than
real
roads
(e.g.,
Bella,
2005,
2008;
Bella
et
al.,
2007;
Boer
et
al.,
2000;
Törnos,
1998).
For
example,
Bella
(2008)
in
a
study
of
speed
profiles
on
11
sites
on
a
two-lane
rural
road
near
Rome,
has
shown
a
relative
validity
between
real
data
and
driving
simulator
data.
Absolute
validity
was
obtained
for
nine
sites.
Higher
speeds
in
simulated
sites
were
found
at
two
sites,
which
appeared
to
be
caused
by
a
lower
perception
of
risk
on
the
simulated
road
than
on
the
real
road
(Bella,
2008).
These
differ-
ences
of
amplitude
in
the
effects
or
absolute
value
(e.g.,
driving
speed)
raise
the
problem
of
extrapolating
the
results
from
virtual
reality
to
real
life.
In
the
case
of
Philip
et
al.
(2005),
it
is
impossible
to
extrapolate
their
results,
except
perhaps
on
group
level,
because
of
the
type
of
driving
simulator
used
(Divided